A guide to what consumer sleep trackers actually measure well, which device class answers which question, and how to tell when the tracking itself has become the problem. For people who already own a ring, watch or under-mattress sensor — or are about to buy one.
Executive summary
Consumer sleep trackers are reasonably good at the boring things: when you went to bed, roughly how long you slept, your resting heart rate overnight, your breathing rate at rest. They are mediocre to poor at sleep staging, and the single-night "sleep score" is a proprietary composite with no published validation behind it in most cases.
The practical consequence is that the right device depends entirely on the question you’re asking. Building a consistent schedule needs very little from a device. Investigating possible sleep apnoea needs a screening prompt and then a clinician — no wearable diagnoses apnoea. Managing training load needs multi-day trends in heart rate and respiratory rate, not last night’s readiness number.
And for a meaningful minority of readers, the honest recommendation is to stop tracking. There’s a clear exit criterion further down.
What we know
Before I wrote about sleep, I ran overnight polysomnography studies. Sleep stages are defined by electrical patterns in the brain. That is the definition.
A ring or watch has access to none of that. It has movement, an optical pulse sensor, usually skin temperature, sometimes blood oxygen estimates. From those, it infers sleep. An under-mattress sensor has even less contact — it reads movement, and sometimes ballistic signals from your heartbeat and breathing through the mattress.
The American Academy of Sleep Medicine set out its position on consumer sleep technology in 2018: these devices are not equivalent to validated clinical tools, should not be used to diagnose or treat sleep disorders, and their claims are frequently unsupported by peer-reviewed evidence. The AASM also allowed that they can be genuinely useful for engaging people in their own sleep behaviour.
What the evidence says
Chinoy and colleagues’ 2021 study in Sleep tested several commercial sleep-tracking devices against simultaneous polysomnography in healthy adults. The broad pattern: the better devices performed roughly comparably to research actigraphy when distinguishing sleep from wake, while sleep-stage performance was weaker and varied considerably between devices.
Consumer trackers generally have higher sensitivity for sleep than specificity for wake: timing and total sleep can be useful trend estimates for many sleepers, but quiet wakefulness is commonly misclassified as sleep, which can overestimate total sleep and miss awakenings. Stage estimates remain weak because the signal a stage is defined by isn’t available at the wrist.
Optical heart rate at rest is one of the more trustworthy numbers your device produces. Respiratory rate at rest is similarly stable. Heart rate variability is noisier and more sensitive to sensor position and averaging — useful as a trend, unreliable as a nightly figure.
A sleep score is a company’s weighted blend of its own estimates, usually undisclosed and periodically revised. There is generally no published validation of the score itself.
Some devices carry regulatory clearance for features that flag possible signs of sleep apnoea across multiple nights. Clearance for a notification is not a diagnosis. AASM guidance is clear that diagnosis requires polysomnography or a home sleep apnoea test, interpreted clinically.
What remains uncertain
Validation studies describe firmware that may no longer exist. Algorithms are updated silently, sometimes substantially, and a device that scored well in a 2021 comparison is not necessarily the same device on your finger now. This is the single biggest reason I don’t publish per-model accuracy rankings.
Most accuracy research is conducted in healthy sleepers in labs. Performance in people with insomnia, untreated apnoea, shift-work schedules, chronic pain or in older adults is thinly studied. There’s also reason to think optical sensors perform unevenly across skin tones and perfusion states; how much that affects overnight sleep outputs specifically is not well characterised.
And the outcome question is genuinely open: we don’t know whether tracking sleep, over years, makes anyone sleep better.
Mechanism
Why can total sleep time still be useful as a trend while staging fails? Because lying still with a low, steady heart rate is a decent proxy for being asleep. It still overestimates sleep when quiet wakefulness looks like sleep.
Stages don’t work that way. REM has a fairly distinctive autonomic signature, so a device can make a reasonable guess at it. But the boundary between light and deep sleep is drawn on brain-wave criteria that leave almost no fingerprint at the wrist. So "deep sleep: 54 minutes" is an estimate derived from an estimate, presented to the minute. The precision of the display is not evidence about the precision of the measurement.
That gap between displayed precision and actual confidence is where most tracker anxiety lives.
Practical implications
Start with the question, not the device.
If the question is "is my schedule consistent?"
You need almost nothing. A phone or any device that records bed and wake times will do. Timing and total sleep can be useful trends across two weeks, but quiet wakefulness may be counted as sleep. Ignore everything else on the screen. If your wake time is drifting, the fix is usually light and morning routine, not hardware.
36-month subscription burden: $0. Hardware: $0 if you use a phone you already own.
If the question is "do I stop breathing at night?"
Lead with clinical evaluation if you have witnessed breathing pauses, loud habitual snoring, morning headaches or daytime sleepiness. That symptom picture needs a sleep study. If you choose a consumer device, limit it to a feature specifically regulator-cleared for apnoea-risk notification in your market. Snore audio may provide optional context for an appointment, but raw single-night oxygen dips, snore scores and the absence of an alert must not be used to rule apnoea in or out. A flag means get tested; no flag does not rule apnoea out.
36-month consumer subscription burden for the recommended clinical-first route: $0. Consumer hardware: $0; clinical testing is separate.
If the question is "am I recovering from training?"
Choose a well-fitting watch that provides overnight resting heart rate, respiratory rate and HRV without a paid tier. Read HRV as a rolling multi-day baseline rather than a nightly verdict. A resting heart rate several beats above your own baseline for a few consecutive nights is worth noticing. One elevated night after a hard session or a glass of wine is just physiology.
36-month subscription burden: $0. Hardware: purchased separately, with price varying by model.
Outputs to ignore
The single-night sleep score. Deep sleep minutes. REM percentage compared against population "averages". Precise awakening counts. "Sleep age". Single-night blood oxygen dips treated as self-diagnosis.
The buy-nothing option
If you’re currently sleeping badly and shopping for a tracker to understand why, a two-week paper log of bed time, wake time and how rested you felt costs nothing and is better evidence for a clinician than a screenshot of a score.
Limitations
This analysis can’t rank specific models by accuracy, because published validation lags firmware and because the studies that exist test particular device versions in particular populations. It can’t tell you whether your device’s apnoea feature will catch your apnoea. And it can’t settle whether tracking helps or hinders any individual — the outcome evidence isn’t there. What it can do is tell you which outputs rest on measurable signals and which rest on inference stacked on inference.
Conclusion
Sleep trackers are better instruments than their critics say and much worse oracles than their marketing implies. Trust timing and cardiorespiratory trends. Treat stages as a rough sketch and the score as a mood, not a measurement. Treat an apnoea flag as a reason to book an appointment; no flag does not rule apnoea out.
And if the number has started arriving in your head before the felt sense of rest does — that’s the signal that matters most. You don’t have to decide tonight. But you’re allowed to take the thing off.
Reader questions
Frequently Asked Questions
The questions that come up most often, answered plainly.
Are the sleep stages on sleep trackers accurate?
They are rough estimates, not minute-accurate measurements. The evidence and mechanism are explained above.
Do sleep trackers detect sleep apnoea?
Some regulator-cleared features can flag risk, but they do not diagnose it. A flag warrants clinical evaluation; no flag does not rule apnoea out.
Ring or watch for sleep tracking — which is better?
Choose the form you can wear comfortably and consistently. Fit matters more than spec; a device you take off at 2am measures nothing.
Should I stop using my sleep tracker?
If the score changes how you feel or plan your day, or makes you more anxious in bed, take the tracker off for two weeks. Removing the scoreboard is a legitimate intervention, not a failure.
Is a sleep tracker subscription worth it?
For the three jobs above, the recommended routes need no paid subscription. Hardware costs remain separate.
Sources
- Consumer Sleep Technology: An American Academy of Sleep Medicine Position Statement (Journal of Clinical Sleep Medicine, 2018)
- Chinoy et al., Performance of seven consumer sleep-tracking devices compared with polysomnography (Sleep, 2021)
- Baron et al., Orthosomnia: Are Some Patients Taking the Quantification of Sleep Too Far? (Journal of Clinical Sleep Medicine, 2017)
- Kapur et al., Clinical Practice Guideline for Diagnostic Testing for Adult Obstructive Sleep Apnea (American Academy of Sleep Medicine, 2017)
